High-dimensional FSI system and Low-Dimensional Modelling Marek Morzyński Witold Stankiewicz Robert Roszak Bernd R. Noack Gilead Tadmor Overview Elements and of High- Dimensional Aeroelastic System Loosely coupled aeroelastic system Computational aspects Elements of the system Solutions ROM with moving boundaries and ALE ROM in design and flow control ROM for AE – sketch of challenges and ideas ROM AE model - motivation Need of ROM in design AIAA 2008, Rossow, Kroll Aero Data Production A380 wing Need of online capable ROMs in feedback flow control 50 flight points 100 mass cases 10 a/c configurations 5 maneuvers 20 gusts (gradient lengths) 4 control laws ~20,000,000 simulations Engineering experience for current configurations and technologies ~100,000 simulations Aeroservoelasticity Aeroelastic control (Piezo-control of flutter, wing morphing, smart structures) MicroAerialVehicles (maneuverability) High- Dimensional Aeroelastic System – ROM testbed t=t+t Tau Code Flow code Fluid forces In-house and AE tools convergence no Deformed CFD mesh, velocities Interpolation Forces MF3 (in-house), Calculix, Nastran yes CFD mesh deformation Interpolation Structural code Structure displacements and velocities Spring analogy Computational aspects – Euler code Mesh: 10 mio elements t=t+t t=80s CPU Power: 16 cores Flow code Fluid forces t=10s convergence no Deformed CFD mesh, velocities Interpolation Forces t=4s / 50s yes CFD mesh deformation t=30s Interpolation t=10s Structural code Structure displacements and velocities One iteration time: 134s (full CSM) / 180s (modal CSM) Computational aspects - RANS Mesh: 30 mio elements (1 mio: surfaces) t=t+t t=400s CPU Power: 32 cores Flow code Fluid forces t=90s convergence no Deformed CFD mesh, velocities Interpolation Forces t=4s / 50s yes CFD mesh deformation t=220s Interpolation t=90s Structural code Structure displacements and velocities One iteration time: 850s (full CSM) / 804s (modal CSM) High-fidelity CFD and CSM solvers CFD - TAU CODE • Finite volume method solving the Euler and Navier-Stokes equations • hybrid grids (tetrahedrons, hexahedrons, prisms and pyramids) • Central or upwind-discretisation of inviscid fluxes • Runge-Kutta time integration • accelerated by multi-grid on agglomerated dual-grids • miscellaneous turbulence models • Parallelized with MPI • Parallel Chimera grids From DLR TAU-code manual CSM MF3: in-house CSM Tool • Finite Element-based • Rods, beams, triangles (1st / 2nd order), membranes, shells, tetrahedrons (1st / 2nd order), masses and rigid elements • Static analysis • Transient (Newmark scheme) • Modal analysis • MpCCI and EADS AE interfaces ALE - Motion of boundary and mesh canonical domain u u u p 1 Re Eulerian approach u = 0 M x C x K x F ( x, x , x , t ) Lagrangian approach Arbitrary Lagrangian-Eulerian (ALE) binds the velocity of the flow u and the velocity of the (deforming) mesh ugrid. For incompressible Navier-Stokes equations the mesh velocity modifies the convective term: With boundary conditions: The fluid mesh can move independently of the fluid particles. Coupling requirements Alenia SMJ CFD N-S hybrid grid with 1.3 mio nodes and 4.7 mio elements (cells) Alenia SMJ FEM model with 2,815 nodes Aerodynamic mesh 12437 nodes Structural mesh 212 nodes Z X Y Pressure forces interpolation Coupling tools The meshes are non-conforming • different discretization • different shape (whole wing/ torsion box only Non-conservative interpolation Conservative interpolation Coupling tools • MpCCi (Mesh-based parallel Code Coupling Interface), developed at the Fraunhofer Institute SCAI • AE Modules, developed in the framework of TAURUS • In-house tools, based on bucket search algorithm AE Modules by EADS and in-house modules perform better in the cases, when only torsion box of the wing was modelled on the structural side. Dynamic Coupling: time integration General aeroelastic equations of motion : [M] x’’ (t) + [D] x’ (t) + [K] x (t) = f (x, x’, x’’, t) Inertial forces Newmark direct integration method xi+1 = xi + t xi‘ + t2 [ ( 1/2 - ) xi‘‘ + xi+1‘‘ ] Damping forces Elastic forces Aerodynamic forces Structural forces Integration in time in CFD (or CSM) code xi+1‘ = xi ‘ + t [ ( 1 - ) xi‘‘ + xi+1‘‘ ] xi+1 = xi + t xi‘ + t2/2 xi‘‘ xi+1‘‘ = ( [M] + t/2 [D] ) -1 { f i+1 - [K] x i+1 - [D] ( xi‘ + t/2 xi‘‘ ) } xi+1‘ = xi‘ + t/2 ( xi‘‘ + xi+1‘‘ ) NEWMARK explicit scheme with = 0 and = 0.5 Fluid mesh deformation • Spring analogy • All edges of tetrahedra are replaced with springs (torsional, semi-torsional, ortho-semi-torsional, ball-vertex, etc.) • The stiffness km of each spring may be constant, or related to element size or distance from boundary • Shephard interpolation (Inverse Distance Weighting) Based on the distances di between a given mesh node and boundary nodes: • Another possibilities: Elastic material analogy, Volume Splines (Radial Basis Functions),Transfinite Interpolation I22 and I23 airplanes from: wikimedia Flutter analysis for I-23 airplane Mach number: Atmospheric pressure: Reynolds number: Angle of attack: Time step: Singular input function: M = 0.166, 0.2, 0.3, 044 P = 0.1 MPa Re = 2e+6 α = 0.026 dt = 0.01 s Fz = 2000 N in time t = 0.01 s Flutter analysis for I-23 airplane Simulation: flutter at Ma=0.44 Experiment: flutter at Ma=0.41 Time history for displacement and rotation in control node on wing Flutter Laboratory IoA and PUT experiment and computations • Scale : • Length - 1:4 • Strouhal number 1:1 Experimental configurations • 5 cases – mass added - 50 grams on the wing's tip - 20 grams in the middle of ailerons - 30 grams on vertical stabilizer + 20 grams on tail plane aileron - 20 grams on horizontal stabilizer - configuration FSI test case 1 #1 - 50 grams on the wing's tip Results of test case 1 #1 - 50 grams on the wing's tip Low-Dimensional FSI algorithm t=t+t Flow ROM Pressure yes convergence no Deformed CFD mesh, velocities Interpolation Forces on structure CFD mesh deformation Interpolation Structural code Structure displacements and velocities Amplitudes of „mesh” modes Reduced Order Model of the flow u u u p 1 Re u = 0 1. GALERKIN APROXIMATION N u [N ] u0 a i ui i 1 2. GALERKIN PROJECTION u ,u i [N ] 0 3. GALERKIN SYSTEM a i 1 Re N l j0 N ij aj N q j0 k 0 ijk a jak Navier-Stokes Equations Projection of convective term u ( u u grid ) u p 1 Re u = 0 Arbitrary LagrangianEulerian Approach NG u grid 1. DECOMPOSITION i 1 ai ui G G 2. GALERKIN PROJECTION u i , ( u u grid ) u u i , u N NG N q ijk a j a k j0 k 0 G G uk G q ijk a j a k j 1 k 0 q ijk u i , u j G N u u i , u grid u ROM for a moving boundary NACA-0012 AIRFOIL 2-D, viscous, incompressible flow = 15˚, Re = 100 (related to chord length) displacement of the boundary and mesh velocity: where: T = 5s and Y1 = 1/4 of chord length DNS with ALE Inverse Distance Weighted First 8 POD modes: 99.96% of TKE ROM for a moving boundary Eulerian ROM vs ref. DNS Dumping of oscillation typical for sub-critical Re ALE ROM vs DNS The first two modes AE mode basis for a flow induced by structure deformations • Test-case: bending and pitching LANN wing • Fluid answer to separated, modal deformations (varying amplitudes) • Fluid answer to combined deformation Pressure field and structure deformation (high-dimensional AE) LANN wing structure ROM AE: CFD → CSM Coupling • We preserve full-dimensional CSM and existing AE coupling tools to interpolate fluid forces on coupling - “wet” - surface; (similarly to Demasi 2008 AIAA) Neighbour search: ae_modules f_cfd2csd Pressure interpolation: ae_modules b_cfd2csd where si (i=1..15) is a distance from CFD node to closest CSM elements • High-dimensional fluid forces retrived from the Galerkin Approximation ROM AE: CSM → CFD Coupling and CFD mesh deformation • Linear CSM: deformation decomposed onto mesh modes; Galerkin Projection of ALE term is performed during the construction of GM • Solution of resulting Galerkin System requires only the input of mesh mode amplitudes NG u grid i 1 ai ui G G q ijk u i , u j G G uk • Time stepping: the mesh deformation/velocity calculated for next time step with the Newmark scheme ui+1 = ui + t ui‘ + t2 [ ( 1/2 - ) ui‘‘ + ui+1‘‘ ] ui+1‘ = ui ‘ + t [ ( 1 - ) ui‘‘ + ui+1‘‘ ] Mode interpolation Parametrized Mode Basis (Reynolds number here) OPERATING CONDITIONS II POD modes time-avg. solution =0.25 =0.50 shift-mode =0.75 Eigenmodes steady solution M. Morzynski & al.. Notes on Numerical Fluid Mechanics 2007 OPERATING CONDITIONS I Tadmor & al. CISM Book 2011 -fast transients Results and Conclusions Advanced platform for FSI ROMs testing open for common research Computations ongoing Treatment of CSM - evolution Linear CSM model Non-linear CSM model Tadmor & al. CISM Book 2011 – control capable AE model Mode parametrization CFD/CSM Coupling Canonical computational domain Coupling in Low-Dimensional AE • Full-dimensional CSM • The algorithm essentially the same as the high-dimensional one • Interpolation of pressures/forces required • Interpolation of boundary displacements and mesh deformation required: dependent on the chosen approach of boundary motion modelling (acceleration forces / actuation modes / Lagrangian-Eulerian / …) – Tadmor et al., CISM book • Modal CSM • The aerodynamic forces on the surface of structure might be related to the POD (or any other) decomposition of pressure field • Thus: interpolation of pressures/forces not required • Mesh deformation (velocity) modes / actuation modes calculated in relation to the eigenmodes of the structure • The amplitudes of „mesh” modes calculated from the amplitudes of eigenmodes of structure (time integration?) • Thus: interpolation of boundary displacements and mesh deformation not required